Syntax - a multi-perspective introduction

[Space Dracula]

Fancy title, isn't it?

In any case, Miekko has volunteered to help me out with this. We've pretty much agreed to approach it on a topic-by-topic basis, analyzing phenomena from different angles where possible. The focus, though, is going to be LFG, since that's what Miekko and I both are most proficient with.

If anyone would like to help out and offer different perspectives, that would be appreciated; particularly things that are significantly different from these approaches.

Lastly, if you feel like buying an LFG book, here are some suggestions:

Intro -- Kroeger's Analyzing Syntax

Real LFG -- Bresnan's Lexical-Fuctional Syntax.

Bresnan was one of the founders of LFG, by the way; she knows what she's doing.

I'll be using the approach of the first book, but much of the material of the second book (which is more formal and advanced, and has better examples) in explaining all this.

[Miekko]

A little intro
Right. Modern syntactical research originates with some rather chomskyan notions, which probably overestimate how central a role syntax has in speech production.Nevertheless, the Chomskian revolution has gotten us a load of good stuff, in that this stuff's actually been _researched_ at all.

Syntax, as I've mentioned earlier, isn't just SVO/VSO/SOV, Gen-N or N-Gen, etc. There's a load of interesting things that go under syntax, including

• ergativity

• quirky case

• voice

• how to figure out where various words like reflexive pronouns and other anaphora refer back to

• how arguments can be omitted, but still understood implicitly

• the concepts of 'topic' and 'focus' are dealt with in syntax, and what kind of stuff different languages permit to be done with them

• why English doesn't permit constructions like "the my car", "his a house", "the that house"

Different theories present somewhat different solutions to these different topics. Regardless of which theory ultimately is closest to the correct answer, I suspect most of them might be useful for conlanging purposes.

What, except for the usual linguistical descriptive purposes, is syntax researched then? Well, syntax is one of the things we can formalize pretty well, and various things about it might tell us some things about how the mind handles language.

Compare the following phrases:
the dog the rat bit barked
*the dog the rat the cat ate bit barked

Why is the other phrase so awkward?
It can be parsed, correctly, to mean 'the dog that was bitten by the rat that the cat ate barked', without any larger difficulties. This suggests something about that kind of nesting and how the brain handles it.

Now, more of that later. We'll probably start out with defining the concept of 'constituents' for you all. And get back to interesting examples like the recently mentioned nesting a bit later.

[Space Dracula]

One of the foundational topics of syntax covered in every syntax book I have (that's introductory at least) is the notion of the "syntactic constituent": things like nouns, verbs, and phrases. It's possible for a constituent at a certain "level" to be more than one word long, as long as the words are connected together in some way as to form a unit.

Here are a few examples of syntactic constiuents:

Dog;

The dog;

The very big dog.

You should notice that all of these can function grammatically as a single unit. These units we'll enclose with [square braces].

Here's are a couple sentences (roughly) broken into chunks:

[[The short-haired domestic cat] bit [the very large dog]]

[[The man [who I saw yesterday]] likes [to wear Italian suits]]

Notice that "bit" is central to the first sentence, and works a sort of a pivot that everything else spins around. As you know, the thing to the left of "bit" in this case is the subject, and the thing to the right is the object. We would usually refer to such an ordering as being SVO. However, there is much more to it than this.

Other questions we can ask include, "Why does the article precede the adjectives?" and "Why does the object often come before the verb's modifiers?"

In theoretical linguistics, it's simply not enough to describe things and leave them how they are; we must continually ask why.

A good question to ask now is not a why but a how: "How do we tell if a word is a constituent or not?" Linguists have provided us with a number of tests for constiuency.

Here's an example from the Kroeger book:

9)
a. John ran up a big hill.
b. John ran up a big bill.

10)
a. Up a big hill John ran.
b. *Up a big bill John ran.

(Note that a preceding * means a sentence is ungrammatical; a preceding ? means that it may be grammatical to some.)

Why is 10b ungrammatical then? Most English speakers would say it has to do with "up"; this intuition is correct. "Run up" is what's called a phrasal verb: a verb with a set preposition that gives it a unique meaning. The two function as one unit semantically.

The reordering of two similar sentences in the same way to see if one is grammatical is a type of test used often to determine syntactic features. In this case, it's a verbal constituency test.

Another pertinent question is, "Why do we have different word types such as V, N, etc.?" Some linguists, including (recently) Chomsky and his followers, have posited that we really don't. I personally hold similar beliefs (namely that there are underlying features to the different categories), although mine aren't of the Chomskyan camp.

Nonetheless, there are many ways to identify traditional word categories with other constituency tests. I won't get into many of these yet, but your assignment for tonight is to come up with an example or two in line with 9) and 10).

[Junes]

Question: how is LFG related to optimality theory? I ask this because I've read an article by Bresnan (The emergence of the unmarked pronoun, PDF) which made heavy use of OT methods.

[Space Dracula]

Question: how is LFG related to optimality theory? I ask this because I've read an article by Bresnan (The emergence of the unmarked pronoun, PDF) which made heavy use of OT methods.

A lot of LFGers use OT to expand LFG to include phonology. Basically the way they relate is LFG generates possible constructions, and OT restricts them. Something to that effect.

[JohnQPublik]

9)
a. John ran up a big hill.
b. John ran up a big bill.

10)
a. Up a big hill John ran.
b. *Up a big bill John ran.

...your assignment for tonight is to come up with an example or two in line with 9) and 10).

Not exactly illustrating the same principle, but still within the larger vein of "why isn't the second one grammatical?", how about:

1)
a. The boy threw a ball at the girl.
b. The boy threw the girl a ball.

2)
a. The boy threw a ball at the window.
b. *The boy threw the window a ball.

[Space Dracula]

Not exactly illustrating the same principle, but still within the larger vein of "why isn't the second one grammatical?"

That's what I meant to ask for, sorry if there was any confusion.

[/quote]
1)
a. The boy threw a ball at the girl.
b. The boy threw the girl a ball.

2)
a. The boy threw a ball at the window.
b. *The boy threw the window a ball.[/quote]

That's quite good, though like you said not the same thing; it's more the sort of thing I'd address from a more semantic angle. Also notice how the movement illustrated in my example above leaves these sentences grammatical, but changes the meaning entirely:

1)
a. The boy threw a ball at the girl.
b. The boy threw a ball at the window.

2)
a. At the window a boy threw a ball.
b. At the girl a boy threw a ball.

This goes to show that, in fact, there is more to language than syntax.

[Miekko]

Here's a term you might benefit from knowing later on:

theta role
Semantic role - for instance, the agent is the agent regardless of whether it's a nominative subject or an oblique argument in a passive or even something else in some constructions in other languages.

[Delalyra]

Yay!

/relurks

[Space Dracula]

Here's a term you might benefit from knowing later on:

theta role
Semantic role - for instance, the agent is the agent regardless of whether it's a nominative subject or an oblique argument in a passive or even something else in some constructions in other languages.

I should also note that I might use all sorts of wacky, half-made-up terms for this, including "functional role", "semantic role", etc.

Also, the next lesson is going to be about how units combine to form larger units, in a rough way, and will consider lexical categories. After that, I'm going to cover things like passive formation, and other more specific topics.

First, though, a little bit on the formal structure of LFG. Don't worry too much about this; use it as a reference later.

LFG basically is built around mutually-constraining structures. This means that each structure limits what the others can do. Unfortunately it makes very little consideration of phonology (something its integration with Optimality Theory is meant to resolve), but nonetheless the two main structures are "functional structure" (f-structure) and "categorial structure" (c-structure). There's also "argument structure" (a-structure) which is kind of a proxy between f- and c-structures; often a-structure is a part of an f-structure, and is lexeme-specific. Each word has its own a-structure. Most of the interesting things in grammar are described in terms of mismatches or conflicts between f-stucture and c-structure.

F-structure is meant to be a sort of "picture" of how the units of the sentence are connected functionally. It shows things like which lexeme is the subject of a clause. C-structure is how f-structure presents itself in usage. F-structure is mapped to c-structure in different ways, but for our purposes I'm going to illustrate things with phrase structure rules (occasionally my own modified version of them), since I can't represent the neat little arrows Bresnan uses.

To make all of this a little more clear, here's an example of an f-structure, a c-structure, and the rules mapping the two (f- and c-structures taken from Bresnan, 2001; rules are mine):

F-structure

Code: Select all

     +-                              -+
     | PRED   'kill<(f SUBJ)(f OBJ)>' |
     | TENSE  PAST                    |
f,g: |                                |
     |        | PRED 'child' |        |
     | OBJ h: | DEF  +       |        |
     |        | NUM  SG      |        |
     +-                              -+

In f-structures, PRED refers to the semantic nucleus of that level of the structure. Somewhat confusingly, nouns are considered PREDs which are zero-valence. TENSE is of course tense. OBJ is the object of the predicate, as called for in the a-structure of "kill". DEF is whether the noun is definite; NUM is its number. Don't worry about "f,g" yet.

The meaning of thie f-structure is "kissed the child", with an unspecified subject.

Here's the c-structure as a tree diagram:

Code: Select all

     VPf
     /  \
    /    \
   /      \
  Vg       NPh
  |       /   \
  |      /     \
  |     /_______\ 
kissed  the child 

Bresnan (2001) makes heavy use of arrows to show how c-structure ties in with a- and f-structures, which is why I'm going to rely a bit more on phrase structure rules. For those of you already acquianted with Chomskyan syntax, one of the more interesting things about LFG is how the f-structure can map to c-structure without any sort of movement, since there are not any "layers" of trees to map to one another.

Anyway, here's a rule to map the f-structure to the c-structure:

VP => V NP

That's it. Whatever is marked for these slots will go in there and all is in harmony.

Now let's try adding a subject:

Code: Select all

       +-                              -+
       | SUBJ   j:[PRED 'NAMED-Mary']   |
       | PRED   'kill<(f SUBJ)(f OBJ)>' |
       | TENSE  PAST                    |
e,f,g: |                                |
       |        | PRED 'child' |        |
       | OBJ h: | DEF  +       |        |
       |        | NUM  SG      |        |
       +-                              -+

Now we add another rule:

S => NP VP

Giving us:

S => NP VP
VP => V NP

Which produces:

Code: Select all

          S
         / \
        /   \
       /   VPf
      /    /  \
     /    /    \
    /    /      \
  NPj   Vg       NPh
  |     |       /   \
  |     |      /     \
  |     |     /_______\ 
Mary  kissed  the child 

Of course, things get more interesting when we hit Australian languages, which can have c-structures like this:

Code: Select all

        ____S_____
       | | | | | |
      / /  |  \ \ \
     |  |  \  |  | \
     |  |   | |  |  |
     NP AUX V NP NP NP

[finlay]

You lost me at the last little bit... but otherwise most of this I have to go into an exam and do in about half an hour. I still don't know what the hell LFG stands for; I think it's worth explaining jargon when you're writing a guide before you use it.

[Miekko]

LFG is 'lexical-functional grammar'

Also, for those who aren't used to tree grammars, here's some explanations:



the triangle between NPh and 'the child' means that the structure of the phrase 'the child' is left undescribed, because it is uninteresting for the purposes of this example.



Now, Spack is using letters after NP to mark what they refer to. If you have two NPs with the same letter after them, they refer to the same thing:



Everything under a node is basically a constituent - the entire Verb Phrase is a constituent, but so are the individual Noun Phrases under the VP.


This is what the Australian language example'd look like as a good tree.

[finlay]

I'd like to see an example of this australian language. If you don't have one, whyever did you do the tree in the first place? I'm intrigued.

By the way, my exam went alright, but it was pretty easy. We have done stuff like most of this before but the exam was less difficult.

Also I'm wondering why you've written it as VPF and VG, and not just VP and V. It seems you're overcomplicating the scenario and not explaining it.

[Miekko]

Also I'm wondering why you've written it as VPF and VG, and not just VP and V. It seems you're overcomplicating the scenario and not explaining it.

Spack'll have to describe that since he's read more formal LFG than I have as of yet.

As for Australian languages, Warlpiri, Jiwarli, Kayardild and a whole bunch of others have C-structures similar to that. As said, they allow total scrambling of the word order (but they do have a whole lotta congruence going on).

Either you'll have to allow constructions like that, or you'll have to ... God forbid ... CROSS LINES!!!

Right. In syntactic trees, you're never supposed to cross lines. Why it is so, I don't know.

However ... in some analyses, I've seen X-barists have the tense/aspect/mood marker cross lines, sine TAM is supposed to be a phrase of its own which encloses the VP so you get
TP[ T[] VP[ V[ V]]], and then, by magic, T[] is _realised_ as an inflection on V, which, imho, quite clearly is crossing lines. (Unless you posit some kind of congruence...)

OTOH, that way, you can consider the VP a constituent that lacks TAM, which'd explain some kinds of extraction phenomena.

[Miekko]

I've used phpSyntaxTree to generate those graphs,btw.

The following strings produce the trees in the post above:

[VPF [VG kissed] [NPh the child]]

[S [NPj Mary] [VPF [VG kissed] [NPh the child]]]

[S [NPj Mary] [VPF [VG cleaned] [NPj herself]]]

[S [NP] [AUX] [V] [NP] [NP] [NP]]

Now, this kind of bracket notation is equivalent to tree notation, but tree notation is so much easier to read, that I'll keep doing that.

ALSO,
I am thinking of doing a rather orthodox introduction to X-BAR Theory, just because it's a good foundation. I am aiming at a somewhat simplified extended X-bar. Okay. That'll sort of introduce you all to our usage of trees, even tho' we'll deviate from it whenever we feel like it

[Space Dracula]

I'd like to see an example of this australian language. If you don't have one, whyever did you do the tree in the first place? I'm intrigued.

By the way, my exam went alright, but it was pretty easy. We have done stuff like most of this before but the exam was less difficult.

Also I'm wondering why you've written it as VPF and VG, and not just VP and V. It seems you're overcomplicating the scenario and not explaining it.

I'll admit I did leave you hanging on the Auslang business, but that was mostly intentional, to use as tie-in to another lesson.

The phrase Bresnan uses that has that tree diagram is this, and is from Warlpiri (it's also good Walpiri, so don't go thinking this isn't a common word order for them):

Code: Select all

wita-jarra-rlu ka-pala      wajilipi-nyi yalumpu  kurdu-jarra-rlu maliki
small-DUAL-ERG pres-3duSUBJ chase-NPAST   that.ABS child-DUAL-ERG  dog.ABS

This translates to "The two small children are chasing that dog".

However, Bresnan notes that Warlpiri isn't entirely free of phrases: some NPs, for example, occur optionally. But there are no VPs. Sometimes there is also a focus slot in front of the Aux.

<hr />

Also I'm wondering why you've written it as VPF and VG, and not just VP and V. It seems you're overcomplicating the scenario and not explaining it.

Spack'll have to describe that since he's read more formal LFG than I have as of yet.

Basically ther eason why they're VPf etc. is to show how the c-structure connects to the f-structure. It's mostly just an explanatory tool as far as I can tell.

Also, IIRC, Role and Reference Grammar does allow for lines crossing. Let me get out my Heavy Tome of RRG and find an example...

Ah, yes.

Here's an example from Dyirbal:

Code: Select all

                      SENTENCE
                          |
                       CLAUSE
                          |
      PERIPHERY --> ____CORE_________
    ______|________/      |          \
   /      |              NUC          \
  /       |               |           ARG
 ARG      |              PRED          |
  |       |               |            |
  NP     LOC              V            NP
  |       |               |            |
ɖugumbil gambiɽa baŋgul buran balan yaɖaŋgu
   |                |___________|______|
   |____________________________|
woman.ABS mountains.LOC DET.ERG see DET.ABS man.ERG

"The man saw the woman in the mountains"

[Space Dracula]

So anyway, I've covered constituents and in general how to identify them. Constituent identification can be hairy, but don't worry too much about it now. In this post I'm going to cover how constituents combine to form larger constituents and so forth. This is basically what syntax is all about: how constituents combine to express meaning.

Constituent categories are generally language-specific as to their specifics and peculiarities, and not all langauges have all constituent types. For example, as I mentioned above, many Australian languages lack a formal verb phrase.

Part of the argument for there being a VP in English is that the object usually follows with the verb in valid reorderings of sentences, which implies that they make a constituent. This isn't so with Australian languages.

Here are a few familiar types of constituent founds in English:

Noun
Verb
Adjective
Preposition
Noun phrase
Verb phrase
Gerund
Prepositional phrase
Independent clause
Dependent clause

There are numerous tests linguists use to argue whether or not a particular language has a particular category, but they really don't matter for the purposes of a general overview. What categories you choose to include in a conlang are your business; you could even do one combining features of both.

IMHO, categories are bunk. The properties generally assigned to them, as far as I can tell, are broadly reducable to a few axes. Whether word categories exist or not is something that brings up a lot of debate. Chomsky, as of late, is against them. Some argue that things rarely, if ever, show properties of both verbs and nouns; they say the fuzziness is mostly in other areas.

A Big Thing about many theories of syntax is recursion. Chomskyan syntax is based on it. Recursion allows a NP made up of several Ns to operate like an N: it's self-similar on different levels, like a fractal. When a language deviates from this, by not having recursively layered structures, meaning has to be conveyed in some other way, usually morphology or the lexicon.

Here's an example of a highly recursive phrase (Kroeger):



Coordination and subordination arise as a consequence of this. Coordination links the phrases on the same level, whereas subordination links one below the other. Subordinated clauses can't stand on their own in most cases (there might be exceptions; if so, tell me).

Here's an example of coordination (from Kroeger):



Note that the conjunction is on the same level as the two noun phrases.

Here's an example of subordination (also Kroeger):



Note that COMP, the complementizer (which causes the type of subordination called, you guessed it, a "complement clause") is above the rest of it.

Here are two subordinate clauses, written like a school exercize, linked by a coordinating conjunction (created by me; correct me if I'm wrong please):



Your homework is to make a tree for "Brien and Miekko rock because they write about syntax".

Next I'm going to cover types of subordination and related things, and then move on to the fun stuff. If someone wants, I can post an example of Chomskyan minimalism, where all tree levels only have two members.

EDIT: I screwed up and made 'Penang' not be part of an NP. Ignore that.

[Echobeats]

Ternary branching! Heretic!

[Echobeats]

Ternary branching! Heretic!
_________________
Linguistics will become a science when linguists begin standing on one another's shoulders instead of on one another's toes.
?Stephen R. Anderson

I just realised that my post and my signature contradict each other brilliantly.

[Miekko]

Ternary branching! Heretic!
_________________
Linguistics will become a science when linguists begin standing on one another's shoulders instead of on one another's toes.
?Stephen R. Anderson

I just realised that my post and my signature contradict each other brilliantly.

What's worse, though?
Ternary branching or crossing lines?

[Space Dracula]

Ternary branching! Heretic!

Ternary branching is a feature in LFG, not a bug.

[finlay]

Spack, two of your trees aren't loading.

[Ceresz]

Well....

I tried:

Please, correct my mistkaes
~Skurai

[Echobeats]

What's worse, though?
Ternary branching or crossing lines?

Crossing lines, because then you might as well not have a theory.

[Miekko]

What's worse, though?
Ternary branching or crossing lines?

Crossing lines, because then you might as well not have a theory.

Unless you have very strong constraints.